Nowcasting GDP using tone-adjusted time varying news topics: Evidence from the financial press
We extract tone-adjusted, time-varying and hierarchically ordered topics from a large corpus of Dutch financial news and investigate whether these topics are useful for monitoring the business cycle and nowcasting GDP growth in the Netherlands. The financial newspaper articles span the period January 1985 up until January 2021. Our newspaper sentiment indicator has a high concordance with the business cycle. Further, we find newspaper sentiment increases the accuracy of our nowcast for GDP growth using a dynamic fac- tor model, especially in periods of crisis. We conclude that our tone-adjusted newspaper topics contain valuable information not embodied in monthly indicators from statistical offices.
Keywords: Factor models, topic modeling, nowcasting
JEL codes C8;C38; C55;E3
Working paper no. 766
- We extract sentiment from the texts of Dutch financial news and show that this sentiment is a strong indicator of the Dutch business cycle. Sentiment correlates strongly with GDP growth and sentiment turns negative when the economy is in a downturn;
- We combine sentiment with a time-varying and layered topic model to provide policymakers and practitioners with valuable insights into the causes of sentiment swings;
- Sentiment indicators derived from newspaper articles contain valuable information not embodied in monthly indicators from statistical offices. The forecast accuracy of our dynamic factor model increases when including the tone-adjusted topics derived from the newspaper articles.
- The added value of newspaper sentiment articles is particularly strong during periods of large declines in GDP;